Test for incompatible HEAP transforms in acorn-optimizer

[RReverser]

While working on #24295, I noticed I could accidentally prevent ASan and SAFE_HEAP from finding any HEAPxx[n] accesses from JS when ALLOW_MEMORY_GROWTH is enabled, and no tests would catch it, because no tests verify that acorn-optimizer passes which change HEAPxx[n] into something else can work in conjunction with each other.

After adding the test, I found that an existing SUPPORT_BIG_ENDIAN feature was already broken in the same way, because it transforms HEAPxx[n] accesses into DataView methods, and since it was running before ASan and SAFE_HEAP, those latter transforms became no-ops. I fixed that in the same PR by moving the big-endian transform to the end.

In the future we might want to consolidate those transforms in a more reliable way (e.g. only run a single code transform that does HEAPxx[n] -> __loadHeap/__storeHeap where those functions do all the work using #ifs), but at least for now this PR fixes an immediate problem and catches potential future regressions.

[RReverser]

Actually, if I add -pthread, there are even more combos that currently unexpectedly succeed... changing into draft until I fix them too.

[RReverser]

Ah yeah, looks like ALLOW_MEMORY_GROWTH doesn't work correctly with ASan / SAFE_HEAP for the same reasons even before my change in #24295.

[kripken]

Perhaps a single test with all 3 features enabled is enough?

[RReverser]

It would be enough to reproduce the issue, but I found that having separate tests helps to narrow down which particular transform is at fault. Just really putting "unit" into the "unit tests".

[kripken]

I guess it's not surprising some of these combinations don't work, given the possible interactions, and that we don't have exhaustive testing...

I'd say that only memory growth is a high priority, of the three (asan is run locally, and devs can work around feature combinations, and big-endian likely has very few users).

[RReverser]

I'd say that only memory growth is a high priority

Yeah I agree, combo of memory growth + ASan is the first one I noticed and then thought it's worth checking other combinations too.

I'd like to merge #24291 + #24295 + one more upcoming PR for simplifying ASan #24314 first, and then I'll fix these combo interactions as it will be somewhat easier to do on top of simpler transforms.